70% of Indian Banks Deploy AI in Production, But Security and Data Usability Block Scaling: Zeta Survey

Indian banks move AI into production, but scaling remains a challenge: Zeta

Indian banks are now deploying artificial intelligence in production across various functions. Scaling these AI deployments faces significant hurdles related to security and data usability. Retail lending and customer service show the biggest operational impact from current AI initiatives. Banks are preparing to use AI in more consequential areas like credit risk and fraud detection. Indian banks have moved beyond experimenting with artificial intelligence (AI), with most institutions now deploying AI in production. However, scaling these deployments across the organisation remains a challenge due to concerns around security, data usability, governance and skills, according to a Zeta survey. Zeta's 2026 CXO survey, based on responses from 40 CXOs across 18 banks and NBFCs, found that 70% of chief data officer respondents place their institutions at either selective or scaled AI deployment, including 30% at scaled deployment. Also Read: India tops global AI adoption with highest share of 'frontier professionals': Microsoft AI adoption is strongest in areas where outcomes can be reviewed and existing controls can contain risks, including customer service, fraud and risk analytics, document processing and software testing. Its integration into end-to-end workflows and consequential decisions remains at an earlier stage. The survey highlights a gap between proving that AI works in production and deploying it repeatedly across an institution without rebuilding data, integrations and controls for every new use case. Retail lending sees biggest AI impact Retail lending emerged as the area seeing the biggest operational impact from AI, with 88% of COOs surveyed identifying it as a meaningful area of impact, followed by customer service at 75%. CASA and back-office operations were cited by 63% each. Zeta said roughly four in 10 CDO respondents were unable to identify a high-ROI AI use case within their institution. However, lack of ROI clarity was rated the lowest barrier to adoption, while security and data privacy emerged as the biggest concern. Data availability isn't the problem About 80% of CIOs and CTOs surveyed described their data environment as mostly ready for AI at scale, although none considered it fully ready. The bigger constraints are around making data usable for AI. About 61% cited insufficient labelled or training data, 53% pointed to privacy and consent and 46% cited siloed data. At the same time, 67% of respondents said their banks were using or piloting AI to enhance or enrich their data. AI adoption slows when it moves from producing to executing AI has also gained ground in software engineering. Around 80% of CIOs and CTOs said their organisations use AI for testing and quality assurance, while 60% use it for code generation. Adoption drops to 40% for code review and 30% each for specifications and documentation, deployment and CI/CD, and incident detection. Also Read: AI cyber risk is biggest immediate threat to global financial stability: FSB chair Andrew Bailey Security and data privacy scored 3.89 out of five as a barrier, compared with 2 out of five for lack of ROI clarity. Banks are now preparing to use AI in more consequential areas. At least 60% of chief risk officers surveyed identified AI-led credit-risk models, predictive early-warning systems and real-time fraud decisioning as priorities over the next 18-24 months. Around 60% said responsible AI frameworks were under development, although none reported organisation-wide implementation. Only 20% described model-risk management as very mature. AI could change work before workforce size Half of operations leaders surveyed expect AI-driven productivity gains to free up capacity for redeployment into higher-value work, while none expect workforce reductions above 20%. Banks are currently building AI capabilities faster through specialist hiring and external partners than internal development, which received the lowest capability score in the survey. "Indian banks have shown that AI creates value in production. The next challenge is making that success repeatable, and the survey is clear about what stands in the way: not conviction, but control," said Sivaram Kowta, President, Zeta India. Zeta said the next phase will depend on banks building shared data, infrastructure, governance and control capabilities that allow successful AI deployments to be replicated across the organisation.

Zeta Survey: 70% of Indian Banks Run AI in Production, but Security Concerns Delay Scale

Zeta today announced the findings of its 2026 CXO survey on the state of AI in Indian banking, based on responses from 40 CXOs across 18 leading banks and NBFCs. AI is now in production at most institutions: 70% of CDO respondents place their banks at selective or scaled deployment, including 30% at scaled deployment. Adoption is strongest in bounded, reviewable areas such as customer service, fraud and risk analytics, document processing and software testing. Integration into end-to-end workflows and consequential decisions is at an earlier stage. The survey points to a clear divide between piloting AI successfully and deploying it repeatably at scale. Banks have proven that AI works in production. What remains harder is reproducing that success across the institution without rebuilding data, integrations and controls for every new use case. The technology estate is more connected than ever, but the capabilities that make it usable by AI, from permissioned data and AI-operational infrastructure to engineering controls, governance and skills, are developing at different speeds. Investment reflects this. Most institutions surveyed direct less than 10% of new-project technology spend to AI, including some with AI across multiple functions. The survey suggests this is not a lack of conviction: lack of ROI clarity is the lowest-rated barrier, and executive scepticism and employee resistance rank below skills and security. Banks are measured because control, not appetite, sets the pace. Key findings from the survey * AI is creating meaningful operational impact, but it remains concentrated in structured workflows. 88% of COO respondents identify retail lending as an area where AI is delivering meaningful impact, followed by customer service at 75%, and CASA and back-office operations at 63% each. Adoption is strongest where work is high-volume and structured and outputs can be reviewed within existing controls. Redesigning end-to-end workflows around AI, defining what it executes, recommends or escalates and where human judgement stays decisive, is at an earlier stage. Roughly four in ten CDO respondents cannot yet name a high-ROI use case in their own institution. * Banks are confident about data availability; making it usable for AI is the harder problem. 80% of CIOs and CTOs describe their data environment as mostly ready for AI at scale, but none consider it fully ready. The constraints they cite are about usability rather than quality: 61% point to insufficient labelled or training data, 53% to privacy and consent, and 46% to siloed data. The gap is less about whether banks hold data and more about whether its meaning, permissions and freshness are available to AI without a separate exercise for every use case. Banks are already using AI on the problem: 67% are using or piloting AI to enhance or enrich their data. * Technology assets are connected; making them operable by AI is the next step. Real-time data platforms and API-first architectures report adoption of 79%, with core modernisation and cloud at 64%. Advanced analytics and MLOps, the capabilities needed to deploy, manage and observe AI workloads repeatedly, stand at 43%. The survey suggests the next infrastructure challenge is not cloud adoption itself but making connected environments able to run AI securely, reliably and consistently at scale. * AI has a foothold in software engineering; adoption thins as AI moves from producing to executing. Among CIOs and CTOs, 80% report using AI in testing and QA and 60% in code generation, compared with 40% in code review and 30% each in specifications and documentation, deployment and CI/CD, and incident detection. Security and data privacy score 3.89/5 as the leading barrier, against 2.0/5 for lack of ROI clarity. Banks see the productivity case clearly. Adoption slows where AI would change or execute rather than produce something a person can review. * Banks are preparing to take AI into consequential decisions; governance is developing alongside. At least 60% of CROs identify AI-led credit-risk models, predictive early-warning systems and real-time fraud decisioning as top priorities over the next 18-24 months. Around 60% say Responsible AI frameworks are under development, although none report organisation-wide implementation, and only 20% describe model-risk management as very mature. As AI moves closer to decisions, controls such as AI identity and permissions, policy enforcement and audit are becoming part of the operating architecture rather than a downstream review. * AI is changing work before it changes workforce size; internal capability-building is catching up. Half of operations leaders surveyed expect AI-led productivity gains to release capacity for redeployment into higher-value work, while none expect workforce reductions above 20%. Banks are building AI skills faster through specialist hiring (3.33/5) and external partners (3.0/5) than through internal development (1.0/5), the lowest capability reading in the survey. Two findings capture the tension most clearly: adoption has moved well past experimentation, while the controls needed to scale it with confidence are still being built. From selective production to repeatable scale Taken together, the findings suggest Indian banking has moved past the challenge of taking AI from experiment to production. Production adoption is real but selective, concentrated where the problem is well understood, outcomes can be reviewed and existing controls contain the consequences of error. The question now is whether banks can take what works in these settings and reproduce it across the institution without rebuilding the surrounding data, integrations, controls and engineering practices each time. The survey indicates that this depends on two shared foundations rather than more individual deployments: a core that AI can use, with banking meaning and permissions travelling with the data, and a control layer that establishes what AI may access, decide and execute and keeps a record of it. Until those exist, each deployment remains a one-off, and a measured budget is the rational response. "Indian banks have shown that AI creates value in production. The next challenge is making that success repeatable, and the survey is clear about what stands in the way: not conviction, but control," said Sivaram Kowta, President, Zeta India. "Banks have connected their core systems. The next step is to make them usable by AI, with banking context and permissions built in, and to put in place the identity, policy and audit controls that let risk and security leaders say yes with confidence. Banks that build these foundations once will find that the tenth deployment costs a fraction of the first." The survey frames this progression through four stage gates: prove relevance; prove value and control; make it repeatable; and embed the capability. Banks are already demonstrating the first two in selected areas. The critical transition towards scale is repeatability: reproducing successful deployments through shared capabilities rather than bespoke effort each time.